st/glsl-to-tgsi: Fix handling of csel(bool, vec, vec).

We were throwing an assertion failure across shader-db on nv92.  I'm
guessing this is a regression from !14573.

Cc: mesa-stable
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15540>

ci/crocus: Disable pixmark-piano trace testing.

It hangchecked for me on hsw recently, too.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15832>

ci/iris: Disable pixmark-piano trace testing.

It gave me a spurious hangcheck on
https://gitlab.freedesktop.org/mesa/mesa/-/jobs/20946403, and this trace
causes trouble for a lot of drivers so just disable it on all iris.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15832>

tgsi/transform: Drop a stale comment.

This method returns void.

Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15782>

tgsi/transform: Make tgsi_transform_shader() manage token allocation.

Previously, the caller allocated storage and tgsi_transform_shader() would
emit into that, returning how many tokens it emitted.  All the callers had
to guess at how much storage was necessary, trying not to over-allocate
but also getting enough that you wouldn't (effectively) silently run out
of space.

Instead, make tgsi_transform_shader() do the allocation for you, taking
just a hint of how much space you think you need, and internally double
size when necessary.  Fixes failures on virgl with fp64 since we've added
more fp64 virglrenderer workarounds and its old "XXX: is this enough?"
allocation wasn't any more.

Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15782>

nir_to_tgsi: Fix the address reg mapping for images and SSBOs to match G-T-T.

I missed these in the previous fix to mimic GLSL-to-TGSI address reg
behavior, which r600 relies on.

Fixes: 4bb9c0a28a5f ("nir_to_tgsi: Use the same address reg mappings as GLSL-to-TGSI did.")

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15824>

nir_to_tgsi: Extract const components of atomic counter offsets into Index.

virglrenderer maps atomic accesses to atomic counter declarations using
the .Index field.  We were previously emitting a .Index of 0 for array
accesses, so virglrenderer would emit
atomicIncrement(first_counter[counter_offset+array_index]).  This would
mostly work because hardware doesn't care about the bounds of counter
declarations, but if the first counter was a non-array, then the [] GLSL
emit gets dropped (can't array access a scalar!) and you'd access the
non-array first_counter instead.

Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15824>

r600: Add a helper function for rat_index_mode, with documentation and assert.

We have some special requirements on the address regs of our incoming
TGSI, which nir_to_tgsi violated.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15824>

r600: Stop using ArrayID to look up atomic counters.

Even if you find a matching array ID, you still need to offset by Index
compared to .start, so the ArrayID lookup didn't help.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15824>

zink/kopper: don't use generated include in kopper interface

this causes build race conditions

Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15830>

zink: don't generate VK_ACCESS_SHADER_READ_BIT barrier for vertex inputs

this is both redundant and illegal

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15827>

zink: don't rely on implicit access for generated barriers

compute shaders will get this wrong, so calculate access manually
using binding info

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15827>

zink: add handling for !sync2 in renderpass dependencies

src/dst stage can't be zero without sync2

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15827>

zink: hook up sync2 extension

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15827>

zink: bitcast InterpolateAtOffset offset to fvec

required by spec

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15827>

zink: clamp out partial texels when creating bufferviews

this is an illegal alignment, so clamp the range to the nearest
texel offset since the shader should be hitting the robustness
case for the partial texel

affects:
dEQP-GLES31.functional.texture.texture_buffer.modify.mapbuffer_readwrite.range_size_65537

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15827>

zink: only get swapchain present semaphore on batch flush if not presented

otherwise this has already been present-waited and can just be used whenever

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15820>

zink: only get swapchain present semaphore on batch flush after acquire

otherwise this was a no-op flush from the frontend that should be ignored

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15820>

zink: unset deferred present barrier on flush

don't need to flush this every frame

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15820>

zink: only trigger deferred present barrier if swapchain has acquired

otherwise this was just pointlessly flushed by the frontend and
can be ignored

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15820>

radv: use flush vgt streamout like PAL does.

This uses WRITE_DATA on the ME engine to reset the register, to match what
PAL does on GFX9+.

This fixes
KHR-GL45.transform_feedback_overflow_query_ARB.multiple-streams-one-buffer-per-stream
on zink/radv.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15812>

intel/batch-decoder: Fix uninitialized scalar variables

CID: 1498516
CID: 1498560

Signed-off-by: Vitalii Lomaka <vitalii.lomaka@globallogic.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15685>

zink: don't emit SpvCapabilityStorageImageMultisample for fbfetch

Acked-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15823>

zink: handle multisampled fbfetch

this uses a shader key to flag the fbfetch as per-sample

it doesn't work, but at least it doesn't error anymore?

Acked-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15823>

zink: handle SUBPASS_MS in ntv

for completeness but also maybe fbfetch

Acked-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15823>

intel/dev: Add ATS-M pci-ids

Ref: Bspec 44477
Ref: https://patchwork.freedesktop.org/series/101907/
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Acked-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15814>

nir_to_tgsi: Fix emitting the sample number for non-array MSAA image access.

It's always in .w, rather than being the next component after the
x/y/array index.

Fixes: c6d3fd8c21e0 ("gallium/ntt: Emit sample index when necessary for image load/store.")

Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15825>

virgl: Disable nir_op_ffloor to avoid sending DFLR to virglrenderer.

This means that we send ffract+fsub in place of a normal FLR, but
hopefully virglrenderer can be fixed (or doubles support removed).

Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15825>

radv: introduce new radv_pipeline_stage structure

This is used to store everything for a pipeline stage like the module,
the NIR, the shader arguments etc. This is inspired from ANV.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15794>

radv: add missing multi inclusion define to radv_shader_args.h

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15794>

radv: rework pipeline and shaders creation feedback

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15794>

radv: re-order shader stages directly in radv_create_shaders()

We will have to access pStages::pNext for modules and this will also
allow to rework feedback creation.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15794>

pvr: Remove logic to set vk_device::alloc.

The field is being already set by vk_device_init().

Signed-off-by: Rajnesh Kanwal <rajnesh.kanwal@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15768>

v3dv: fix bogus VkDrmFormatModifierProperties2EXT usage

The array is allocated for VkDrmFormatModifierPropertiesEXT, so
writring entried with type VkDrmFormatModifierProperties2EXT is
bogus.

It seems this was a mistake added with a change intended to get
rid of VK_OUTARRAY_MAKE, that changed the type of the write by
mistake.

Fixes: 56a2ccf05864 ('v3dv: Stop using VK_OUTARRAY_MAKE()')

Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15819>

ci: Disable Link Power Management with RTL8153

There are reliability problems with the RTL8153 ethernet driver under
certain network loads, related to incompatibility of the device with
Link Power Management.

Add usbcore.quirks=0bda:8153:k to the kernel command line to enable the
USB_QUIRK_NO_LPM option.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15791>

Revert "ci/freedreno: Disable a618 jobs"

This reverts commit 96e17287b4abe6ac1e0245473ed9ccfe40d0d54a.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15791>

v3d: do not leak BO on query begin

This happens when we have a sequence of multiple beginQuery / endQuery
with the same target and query identifier.

As a BO is created on beginQuery but not free on endQuery (we need to
wait until either explicitly deleting the query or querying the
results), in the above sequence we are basically leaking the BO.

This ensures that any BO created before is unreference before creating
the new one.

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15796>

wgl: do not disable error-dialogs by default

We don't know if an application developer wants error-dialog by default
or not, even when using a Mesa OpenGL driver.

So let's not assume this is a good thing, and instead make an option for
this behavior that we can enable for CI testing.

Reviewed-by: Louis-Francis Ratté-Boulianne <lfrb@collabora.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15485>

util: limit error-dialogs to win32

These dialogs only exist on Windows, so let's not even expose a util
function for this on other platforms.

The code is only ever called from Windows specific code anyway.

While we're at it, clean up the name a bit as well.

Reviewed-by: Louis-Francis Ratté-Boulianne <lfrb@collabora.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15485>

wgl: rename force-msaa env-var

This env-var is not specific to the SVGA driver, so we should rename it
for consistency.

While we're at it, let's document the feature.

Reviewed-by: Louis-Francis Ratté-Boulianne <lfrb@collabora.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15485>

radv: Disable NGG for GS with suboptimal output vertex count.

When GS has a lot of output vertices: In this case,
the occupancy of NGG GS is very low because API GS invocations
can't even occupy a single Wave32 wave.
Therefore the legacy pipeline performs better here.

Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6260
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15767>

dzn: Support independent depth/stencil access

Needed for VK_KHR_maintenance2. While at it, fix various places where
we were only issuing resource state transition on the first sub-resource
instead of doing it per-layer/aspect.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Fix 2D <-> 3D blits

layer_count == 1 doesn't imply image_is_3d. So let's add explicit
is_3d checks where appropriate.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Lower partial copy of multisample resources to blits

Unfortunately that won't work on transfer queues, but we don't have a
better option right now.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

vulkan/util: Make STACK_ARRAY() C++-friendly

C++ doesn't like the {0} initializer pattern when the first
field is not an integer, let's use the {} when __cplusplus is
defined.

Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Drop extra blank line in dzn_CmdCopyImage2()

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Check image view usage instead of image usage when creating an image view

So we take VkImageViewUsageCreateInfo extension instead of ignoring it.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Force sampleCounts to 1 for bgra4 images

Those are not expected to be used as render-target, and Vulkan mandates
that such formats get their sampleCounts set to VK_SAMPLE_COUNT_1_BIT.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Get rid of dzn_GetPhysicalDeviceProperties()

Rely on the vk_common_ wrapper.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Get rid of dzn_GetPhysicalDeviceFeatures()

And rely on the vk_common_ wrapper to get it implemented.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: 3D array images don't exist

Let's force maxArrayLayers to one in that case.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Set bufferFeatures to zero on depth/stencil formats

Those are not supposed to advertise buffer features.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Make sure the properties are all zero when the format is not supported

Move one of the is_supported() check before we start filling the
structure so we don't end up with a partially filled object when
we return VK_ERROR_FORMAT_NOT_SUPPORTED (which deqp doesn't seem to like,
so it's probably coming from a spec requirement).

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Return a valid imageFormatProperties.maxMipLevels

maxMipLevels is encoding the maximum number of MIP levels, but
dzn_physical_device_get_max_mip_levels() return the maximum MIP level.
Let's rename the function and add one to the returned value to fix
the problem.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Fix 3D <-> 2D image copies

We just need to treat layers as slices when manipulating 3D resources
whose slices are coming from/going to 2D array layers.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Support 2Darray views on 3D images for color attachments

Those are declared as 3D RTVs in D3D12, and layers are treated as
slices.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Fix dzn_image_get_rtv_desc() for 3D views

VK_REMAINING_ARRAY_LAYERS maps to -1 in the D3D12 world. Let's make sure
we set WSize to -1 in that case, because the layer_count calculated by
dzn_get_layer_count() won't work for 3D images which never have more
than one layer (in case of 3D images, we treat slices as layers).

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Replace C++ references by pointers

Let's keep as much as we can in plain C.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Align the default case in dzn_image_view_prepare_dsv_desc()

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Don't crash when EndCommandBuffer() returns an error

Leftover from a debug session.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Fix dzn_translate_viewport() when height < 0

Since negative height is not a thing in D3D12, we need to adjust the
TopLeftY accordingly.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

dzn: Fix pipeline creation when rasterization is disabled

We use some of the VkGraphicsPipelineCreateInfo fields that should
be ignored when rasterization in disabled, assuming those who be set
to NULL by the caller in that case, which is not mandated by the
Vulkan specification.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15698>

virgl/ci: Uprev virglrenderer and crosvm

Use a patch file for crosvm instead of relying on private repositories for faster
uprevs.
Use latest virglrenderer to keep the tests in-sync.

Signed-off-by: Corentin Noël <corentin.noel@collabora.com>
Reviewed-by: Cristian Ciocaltea <cristian.ciocaltea@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15772>

ci: Only apply patches with the build-skqp prefix

Allows to ship patches for other components too.

Signed-off-by: Corentin Noël <corentin.noel@collabora.com>
Reviewed-by: Cristian Ciocaltea <cristian.ciocaltea@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15772>

radv: stop relying on shader modules after SPIRV->NIR

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15766>

radv: stop passing the module to the compiler debug callback

After SPIRV->NIR, the driver shouldn't rely on the module. This will
still report messages via VK_EXT_debug_report but the object will be
NULL.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15766>

radv: drop the module reference for enable_mrt_output_nan_fixup

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15766>

radv: drop the module reference in radv_can_dump_shader_stats()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15766>

radv: drop the module reference in radv_can_dump_shader()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15766>

radv: copy the spirv module for debugging after compilation

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15766>

radv: remove more references to the pipeline layout during compilation

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15749>

radv: lower ycbcr textures just before applying the pipeline layout

This shouldn't change anything.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15749>

radv: assert that the arg is declared when used in get_scalar_arg()

Help debugging.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15789>

radv: add radv_is_vrs_enabled()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15792>

radv: add a new helper to initialize various type of pipelines

This is common to graphics, compute and library pipelines.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15792>

radv: add radv_generate_pipeline_key() for common graphics/compute keys

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15792>

radv: remove unused parameters in radv_get_{wave,ballot_bit}_size()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15792>

radv: use radv_pipeline_has_ds_attachments() more

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15792>

radv: do not check if VkPipelineRenderingCreateInfo is NULL

The driver converts legacy render pass to dynamic rendering, so this
structure should always be in pNext.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15792>

broadcom/compiler: prefer reconstruction over TMU spills when possible

We have been reconstructing/rematerializing uniforms for a while, but we
can do this in more scenarios, namely instructions which result is
immutable along the execution of a shader across all channels.

By doing this we gain the capacity to eliminate TMU spills which not
only are slower, but can also make us drop to a fallback compilation
strategy.

Shader-db results show a small increase in instruction counts caused
by us now being able to choose preferential compiler strategies that
are intended to reduce TMU latency. In some cases, we are now also
able to avoid dropping thread counts:

total instructions in shared programs: 12658092 -> 12659245 (<.01%)
instructions in affected programs: 75812 -> 76965 (1.52%)
helped: 55
HURT: 107

total threads in shared programs: 416286 -> 416412 (0.03%)
threads in affected programs: 126 -> 252 (100.00%)
helped: 63
HURT: 0

total uniforms in shared programs: 3716916 -> 3716396 (-0.01%)
uniforms in affected programs: 19327 -> 18807 (-2.69%)
helped: 94
HURT: 50

total max-temps in shared programs: 2161796 -> 2161578 (-0.01%)
max-temps in affected programs: 3961 -> 3743 (-5.50%)
helped: 80
HURT: 24

total spills in shared programs: 3274 -> 3266 (-0.24%)
spills in affected programs: 98 -> 90 (-8.16%)
helped: 6
HURT: 0

total fills in shared programs: 4657 -> 4642 (-0.32%)
fills in affected programs: 130 -> 115 (-11.54%)
helped: 6
HURT: 0

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15710>

freedreno/a6xx: Fix SP_DS_CTRL_REG0 definition

Bit 20 isn't actually MERGEDREGS, the mode for the entire geometry
pipeline is controlled by SP_VS_CTRL_REG0::MERGEDREGS and it appears to
be something preamble-related instead since writing any register in the
preamble hangs if it's set. This fixes those hangs on freedreno and
turnip since we no longer set it.

Fixes: fccc35c2def ("ir3: Add preamble optimization pass")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15801>

mesa/st: simplify st_can_add_pointsize_to_program iterator

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15699>

mesa/st: don't precompile the pointsize upload variant anymore

this is no longer likely to be used, so precompile the base variant now

also delete some now-unused code

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15699>

mesa/st: only flag pointsize constant uploads if they're needed

now that shaders are guaranteed to have a pointsize export, the only time
the variant using the uploaded constant is needed is when pointsize != 1.0

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15699>

mesa/st: only use constant upload pointsize variants if pointsize != 1.0

it's not that common for apps to need varying pointsize, so now that shaders
are guaranteed to have the export in the shader, the constant version is almost
never used

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15699>

mesa/st: also add pointsize to fixedfunction vertex shaders as needed

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15699>

mesa/st: always inject a 1.0 pointsize for vertex stages

since 1.0 is used in nearly every case, drivers requiring this exporting
can avoid potential shader variants by adding a 1.0 export to the base
shader variant and the only using the ubo upload when pointsize is explicitly
set for wide point functionality

drivers can then be responsible for removing unused pointsize exports
as needed (or desired)

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15699>

mesa/st: declare added pointsize var as hidden

ensure this isn't counted as part of the shader and ignored for e.g.,
glGetProgramInterfaceiv

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15699>

nir/lower_point_size_mov: fix check for overwriting existing pointsize

this should match the comment and allow overwriting injected pointsize
variables regardless of whether xfb is flagged

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15699>

mesa: add a bool indicating when pointsize == 1.0

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15699>

Revert "ci/freedreno: Reduce concurrency when replaying traces on a630"

This reverts commit d948f32365f53aa0677fef833ebaa2cc09d9ab67.

I think that fixing the timeout will have resolved this problem.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15805>

freedreno: Fix the cpu-prep wait to be "infinite".

We don't need to restrict our timeout to 5 seconds, because the kernel's
hangcheck will ensure that the wait completes in finite time if the GPU
gets wedged.  If the GPU is making progress, we don't want to time out
early and have pipe_transfer_map() return an error, causing glReadPixels()
to throw a confusing GL_OOM even though we're not out memory.

The INFINITE arg to this function isn't actually infinite, it's limited to
an hour.  But an hour of GPU processing to wait on is probably plenty.

This 5s timeout has caused problems with the CTS on freedreno at high
parallelism, and I suspect is the cause of recent issues in the closed
traces replay jobs.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15805>

glsl: fix needs_lowering() call in varying packing pass

Here we remove the outer arrays on geom and tess shaders where
needed. Without this the pass can sometimes attempt to pack a
varying on only one side of the shader interface where it is not
actually needed. The result can be mismatching varying types.

Fixes: d6b9202873f0 ("glsl: disable varying packing when its not safe")

Tested-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15761>

docs: update features for lavapipe

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15808>

zink: ci updates

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15808>

zink: add error logging for SRGB framebuffer without KHR_swapchain_mutable_format

this is going to explode, so at least print an error explaining why

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15808>

lavapipe: KHR_swapchain_mutable_format

it Just Works(tm)

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15808>

zink: adds refs to user index buffers when tc is not active

there are no ref tricks to abuse in this case, so add our own ref

fixes #6273

cc: mesa-stable

Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15807>

zink: allow lod for RECT sampler types

the RECT is a lie, so this is fine

Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15804>

nir/lower_tex: avoid adding invalid LOD to RECT textures

this is illegal

Fixes: 74ec2b12be1 ("nir/lower_tex: Rework invalid implicit LOD lowering")

Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15804>